VMM3A ASIC: Detector Signal Processing

• VMM3A ASIC is a 64-channel front-end that integrates amplification, shaping, discrimination, and peak finding for advanced detector readout.
• It supports continuous self-triggered operation with high-rate performance, achieving precise charge and timing measurements in various detector systems.
• The ASIC features flexible calibration and synchronization techniques, enabling optimization for applications ranging from MPGD trackers to neutron detectors and calorimetry.

Searching arXiv for recent VMM3a papers to ground the article. Tool call: arxiv_search The VMM3a ASIC, also written VMM3A in parts of the literature, is a 64-channel front-end application-specific integrated circuit originally developed for the ATLAS muon upgrade and subsequently adopted in a broad set of detector systems, including Micromegas, GEMs, time projection chambers, neutron detectors, and SiPM-based calorimetry. In the literature it is consistently treated as a configurable, self-triggered front end that combines charge amplification, shaping, threshold discrimination, peak finding, and simultaneous charge-and-time digitization, most commonly within the RD51 Scalable Readout System (SRS) or ESS-derived back-end chains (Pfeiffer et al., 2021, Enciu et al., 21 Jun 2026, Bearden et al., 2024).

1. Functional architecture

The VMM3a is described as a 64-channel ASIC whose per-channel signal-processing chain comprises a preamplifier, shaper, discriminator, and peak finder. Reported programmable settings include preamplifier gain in 8 steps from 0.5 to 16 mV/fC and peaking times of 25, 50, 100, and 200 ns; the chip is also described as supporting polarity switching, adjustable threshold discrimination, neighboring logic, and both amplitude and timing measurement (Enciu et al., 21 Jun 2026, Pfeiffer et al., 2021, Flöthner et al., 17 Jan 2025). In one cited description, the ASIC is fabricated in a 130 nm GlobalFoundries 8RF-DM process, reflecting its origin in ATLAS New Small Wheel front-end development (Flöthner et al., 17 Jan 2025).

Across the detector-specific papers, the VMM3a output is characterized in closely related ways. In the GEM-TPC literature it is the device that “provides the peak amplitude and time at peak for signals crossing a threshold level,” while in the sTGC study it provides a discriminator / 1-bit digital output for hit timing together with a 10-bit ADC output for pulse-height measurement (Garcia et al., 16 Jan 2025, Moleri et al., 2021). In the RD51 continuous-readout implementation, each hit is encoded as a 38-bit word containing flag bits, channel number, a 10-bit charge quantity, an 8-bit fine time, and a 12-bit coarse timestamp; the same 38-bit hit-word structure is also described in the calorimetry deployment (Pfeiffer et al., 2021, Bearden et al., 2024).

Several auxiliary functions recur in the literature. The chip includes an internal test pulser for calibration and rate studies, and the Multi-Blade deployment emphasizes the Monitor Output pins, programmable via I$^2$C, for threshold, pedestal, selected-channel voltage level, and internal temperature-sensor readback (Pfeiffer et al., 2021, Piscitelli et al., 2024). These features are significant because much of the published work on VMM3a concerns detector-system adaptation and calibration rather than isolated chip characterization.

2. Readout-system integration

The principal ecosystem for VMM3a in gaseous-detector R&D is the RD51 SRS. In the HYDRA implementation, the readout chain is stated explicitly as detector pads $\rightarrow$ VMM Hybrid $\rightarrow$ DVMM adapter $\rightarrow$ FEC $\rightarrow$ Ethernet $\rightarrow$ DAQ PC, with each hybrid hosting two VMM3a chips for 128 channels on a $5 \times 8\ \mathrm{cm^2}$ board interfaced through a 140-pin Hirose FX10A connector (Enciu et al., 21 Jun 2026). Earlier RD51 rate-capability work describes a closely related hybrid architecture with two VMM3a ASICs, 128 channels total, a Spartan-6 FPGA, HDMI transport to DVMM and FEC stages, and scaling to eight hybrids per DVMM/FEC (Pfeiffer et al., 2021).

The ESS neutron-instrument chain is organized differently but preserves the same front-end role for the ASIC. In the Multi-Blade deployment at PSI, each cassette is connected to one VMM3a hybrid, with 96 of 128 available channels used, and the downstream path is hybrid $\rightarrow$ Front-End Assister $\rightarrow$ Readout Master Module $\rightarrow$ Event Formation Unit $\rightarrow$0 Kafka $\rightarrow$1 FileWriter/Nexus conversion, together with live visualisation tools (Piscitelli et al., 2024). The paper emphasizes that the Amor installation is a downscaled but ESS-equivalent demonstrator for ESTIA, FREIA, and the Test Beam Line, making the VMM3a front end part of an instrument-scale neutron-data pipeline rather than a stand-alone MPGD DAQ.

Synchronization is a recurrent integration problem. In the RD51 SRS literature, a Clock and Trigger Fanout card distributes the common clock needed for continuous self-triggered readout, while the HYDRA implementation specifies a 44 MHz base clock and a custom synchronization path because the VMM3a/SRS chain does not natively support a White Rabbit timestamp receiver (Enciu et al., 21 Jun 2026, Scharenberg et al., 2023). HYDRA addresses this by distributing heimtime and trigger signals through pad-plane adapters, add-on clock buffers, flex PCBs, and spare VMM channels via 500 fF AC coupling, permitting offline reconstruction of global time within the R$\rightarrow$2B DAQ environment (Enciu et al., 21 Jun 2026).

3. Continuous readout, timing, and rate capability

Continuous self-triggered operation is the defining systems property of the VMM3a literature. For HYDRA, the ASIC is reported to operate in continuous readout mode with rates up to 4 MHz per channel without dead time limitation in the operating mode used there, which is presented as crucial for a high-background, high-trigger-rate TPC environment (Enciu et al., 21 Jun 2026). The RD51 rate-capability study places this in a more granular framework: the digitization/conversion time is about 250 ns, setting an intrinsic upper bound near 4 Mhits/s for a single channel, with a practical observed single-channel maximum of 3.6 Mhits/s and an optimized continuous-readout throughput of 8.888 Mhits/s per VMM at the firmware-output level and 20.8 Mhits/s per FEC (Pfeiffer et al., 2021).

Timing performance is likewise reported at several levels. The HYDRA paper describes fast timing of order 1 ns as part of the suitability argument for gaseous-detector readout, whereas the RD51 beam-telescope work states electronics time resolutions between 0.5 and 2 ns and shows that the system enables detector time-resolution measurements better than 10 ns (Enciu et al., 21 Jun 2026, Scharenberg et al., 2023). In the Multi-Blade neutron system, the timestamp is part of each VMM3a hit and is used directly for time-of-flight reconstruction, while the same event record also preserves charge information for weighted-position reconstruction and software filtering against gamma background (Piscitelli et al., 2024).

The literature also makes clear that rate capability is not a property of the ASIC alone. The RD51 rate paper shows that throughput depends on token passing, serialization, buffering, and the number of active channels, and identifies the hybrid/VMM readout side rather than the FEC Ethernet as the main bottleneck in high-rate X-ray imaging (Pfeiffer et al., 2021). The neutron-instrument literature reaches a similar conclusion from a different angle: although the Multi-Blade back-end can sustain large aggregate rates, the VMM3a output pulse length of about $\rightarrow$3–$\rightarrow$4 and the need for about $\rightarrow$5 separation on the same electrodes imply an effective local limit around $\rightarrow$6, or about $\rightarrow$7 neutron events/s for the most likely three-hit event topology (Piscitelli et al., 2024).

4. MPGD and TPC deployments

Within MPGD research, VMM3a has become a standard front end for beam telescopes and tracking detectors. The RD51 VMM3a/SRS beam telescope replaced APV25-based readout in a three-triple-GEM telescope with $\rightarrow$8 active area, $\rightarrow$9 strips at 400 $\rightarrow$0m pitch, and simultaneous readout of detector and scintillator timing signals (Scharenberg et al., 2023). That system exploits threshold-based self-triggered acquisition, neighboring logic, and a modified $\rightarrow$1-weighting scheme for position reconstruction; the reported outcome is spatial resolution around 50 $\rightarrow$2m, time-resolution studies better than 10 ns, and rate capability in the MHz class, with recorded particle interactions saturating at about 1 MHz for minimum-ionizing particles (Scharenberg et al., 2023).

For Triple-GEM readout in the COMPASS++/AMBER context, the VMM3a was evaluated primarily from the standpoint of self-triggering and noise. The published ENC study reports that, at the highest gain and peaking-time settings, the noise behaves approximately as

$\rightarrow$3

and that for matched settings of roughly 4.5–4.6 mV/fC gain, 50 ns peaking time, and 30 pF input capacitance, the VMM ENC is roughly 1000 electrons compared with around 1200 electrons for APV25 (Terzimpasoglou, 2022). The same work treats adaptive gain, discriminator-based hit detection, neighboring logic, and continuous readout as the functional reasons VMM3a is attractive for self-triggered GEM systems.

The GEM-TPC literature uses VMM3a as the enabling front end for two distinct but related TWIN configurations. In the ultra-low-material HGB4-2 detector, two GEM-TPCs share a single gas volume inside one vessel, with one chamber rotated by $\rightarrow$4 relative to the other, and the VMM3a/SRS strip readout digitizes amplitude and assigns a time stamp for each strip signal, allowing track-coordinate reconstruction from drift-time information and strip-cluster position (Garcia et al., 16 Jan 2025). In this configuration the detector operated with beam flux up to 200k particles per spill and collected data uninterruptedly at rates up to 50 kHz over long periods; it also operated stably for three weeks with no destructive discharges. The reported horizontal-plane resolution is about 180 $\rightarrow$5m for Ar/CO$\rightarrow$6 (70/30%) and He/CO$\rightarrow$7 (70/30%), improving to about 135 $\rightarrow$8m for He/CO$\rightarrow$9 (90/10%), while the vertical-plane resolution is approximately 155 $\rightarrow$0m for all gas mixtures. The same paper ties these results to a detector material budget of $\rightarrow$1 (Garcia et al., 16 Jan 2025).

A complementary development is the t0-less TWIN GEM-TPC, in which the self-triggered continuous readout is used to eliminate the need for an external $\rightarrow$2. The analysis combines the absolute drift times from the two mirrored TPCs so that $\rightarrow$3 is removed by subtraction and proper matching satisfies $\rightarrow$4 (Flöthner et al., 17 Jan 2025). The paper reports an upper limit of $\rightarrow$5 from the time-difference proof, and, using reference tracks, residual-based resolutions of $\rightarrow$6 in $\rightarrow$7 and $\rightarrow$8 in $\rightarrow$9, together with horizontal and vertical angle reconstruction (Flöthner et al., 17 Jan 2025).

Pad-readout deployments extend the same front end to resistive MPGD technologies. In a comparative study of Micromegas, RPWELL, and $\rightarrow$0-RWELL with pad readout, all detectors were operated with VMM3a at gain $\rightarrow$1 and shaping time $\rightarrow$2, with thresholds around 10 mV for the tested detectors (Zavazieva et al., 2 Nov 2025). Reported efficiencies are about 97% for Micromegas, about 96% for RPWELL, and about 98% for $\rightarrow$3-RWELL, with time resolutions of about 25 ns, about 25 ns, and about 15 ns, respectively. The continuous self-triggered mode also enabled in-beam and off-beam studies of discharge-like activity and near-breakdown behavior (Zavazieva et al., 2 Nov 2025).

5. Neutron detection and calorimetry

The VMM3a has also been adapted to neutron-instrument front ends. In the Multi-Blade detector commissioned at the Amor reflectometer, the ASIC reads out both positive strip signals and negative wire signals from boron-10-based MWPC cassettes, a polarity flexibility explicitly cited as one reason for its suitability (Piscitelli et al., 2024). The detector signal formation time extends up to about 300 ns, with an average around 150 ns, so the deployed configuration uses the largest shaping time of 200 ns. Operationally, the instrument ran with direct-beam rates of a few MHz over about $\rightarrow$4, corresponding to a few hundred kHz in a single cassette, and the pulse-height spectra did not shift toward lower amplitudes at the higher rate, indicating no detectable space-charge saturation in the tested regime (Piscitelli et al., 2024).

A second neutron application is the T-REX Multi-Grid detector at ESS. There the VMM3A is used as the readout electronics for the wires and grids of the TRP-1 prototype with peaking time set to 200 ns, gain set to 3 mV/fC for anode wires, and 4.5 mV/fC for cathode grids (Backis et al., 2 Oct 2025). The central technical issue is that Multi-Grid charge collection is slow and position dependent, especially near voxel corners, so short shaping can fail to integrate the full charge. Consistent with this, the paper reports that VMM3A pulse-height spectra retain the $\rightarrow$5 bump near blade centers but wash out toward voxel corners, while a CREMAT chain with an effective shaping/peaking time of about $\rightarrow$6 integrates the charge more completely (Backis et al., 2 Oct 2025). Even so, the measured central-region MG/BM count ratios with VMM3A are about 10% higher than CREMAT for $\rightarrow$7 mm, and the averaged measured ratios correspond to 86% of the simulated value for VMM3A versus 58% for CREMAT, a result attributed to count losses in the CREMAT readout chain rather than superior intrinsic integration by the ASIC (Backis et al., 2 Oct 2025).

The first calorimetric use of the VMM3a is reported for the SiPM-based ALICE FoCal-H prototype. Because the native dynamic range proved insufficient for large hadronic-shower signals, the system duplicated each SiPM output into a high-gain and low-gain path on a dedicated adapter board, implementing a 16:1 charge-division coupling scheme and reading 249 detector channels with four VMM hybrids (Bearden et al., 2024). In the high-gain-only configuration, deviation from linearity exceeds 2% at 250 GeV and reaches about 10% at 350 GeV. With the mixed HG/LG reconstruction, the 2% deviation point shifts to 300 GeV; at 200 GeV, replacing saturated HG values with calibrated LG values changes the reconstructed charge by about 4.7%, and the relative resolution changes from 14.8% for HG-only to 16.6% for the mixed reconstruction (Bearden et al., 2024). These results establish that the ASIC can be adapted beyond gaseous readout, but only through explicit system-level compensation of its dynamic-range limits.

6. Calibration, limitations, and detector-dependent optimization

The VMM3a literature places substantial emphasis on calibration and noise characterization. In HYDRA, the electronics noise was estimated with the S-curve method and found to be about 1000 electrons ENC without TPC HV and about 5000 electrons ENC with the TPC operating, leading to the interpretation that the TPC itself dominates the noise rather than the modified VMM readout chain (Enciu et al., 21 Jun 2026). The same study reported adapter-board crosstalk levels of 1.3 mV NEXT and below 0.2 mV FEXT, less than 0.1% of the 3.3 V drive signal, while the Triple-GEM ENC study found a 16% variation among 10 tested VMM chips and stressed the importance of detector-strip and protection-circuit capacitance in determining noise (Enciu et al., 21 Jun 2026, Terzimpasoglou, 2022).

Several publications also document detector-specific limits that are sometimes misconstrued as limitations of the ASIC alone. In the RPWELL study, the 200 ns shaping time integrates only about 30% of the total avalanche charge because the detector signal is much slower than the VMM3a integration window; similarly, the Multi-Grid study shows that short shaping under-collects charge near voxel corners where field gradients are weak and drift is slow (Zavazieva et al., 2 Nov 2025, Backis et al., 2 Oct 2025). These cases do not indicate readout failure in a general sense, but they do show that detector pulse duration can exceed what the available peaking times integrate efficiently.

Saturation and dead-time behavior are another recurring theme. The sTGC emulation study states that sTGC signals can be about 10 times larger than Micromegas signals, so the VMM3a must be configured carefully to avoid saturation and dead time while maintaining high efficiency (Moleri et al., 2021). That work attributes inefficiency under intense gamma background to pileup, jitter, and deep saturation, extracting approximate intrinsic dead times of $\rightarrow$8 ns for $\rightarrow$9 and $\rightarrow$0 ns for $\rightarrow$1, and noting that if 10-bit digitization is active the fixed digitization/re-arming interval rises to about 200 ns (Moleri et al., 2021). In calorimetry, saturation appears directly at $\rightarrow$2 in the HG channel, whereas in the resistive-MPGD pad study the authors conservatively define a discharge-like event as any cluster containing at least one pad above 900 ADC because saturated signals are redistributed above that level after equalization (Bearden et al., 2024, Zavazieva et al., 2 Nov 2025).

The operational literature also points to concrete upgrade directions. For the ultra-low-material GEM-TPC, the unexplained constancy of the vertical-plane resolution motivates improved cluster-timing algorithms, and the authors identify a pixelated readout plane as the next upgrade step for better angular correction and tracking precision (Garcia et al., 16 Jan 2025). In the t0-less TWIN GEM-TPC, field non-uniformity is suggested as a source of $\rightarrow$3-residual asymmetry, and strip-based 3D reconstruction is explicitly limited by an individual VMM3a channel dead time of about 200 ns (Flöthner et al., 17 Jan 2025). The FoCal-H study proposes individual SiPM bias tuning per channel and current monitoring, while the Multi-Grid neutron study indicates that a longer shaping time, around 500 ns, would bring the position dependence closer to the CREMAT level but is not feasible with the current hardware (Bearden et al., 2024, Backis et al., 2 Oct 2025). This suggests that VMM3a is best understood not as a universally optimal front end, but as a highly configurable platform whose effectiveness depends on detector capacitance, signal duration, synchronization architecture, and the bandwidth and reconstruction strategy of the surrounding system.